The Duality AI's Offroad Autonomy Segmentation challenge is a competition designed to explore cutting-edge AI training techniques. Participants will train a robust semantic segmentation model using a provided synthetic dataset of a desert environment. The synthetic data is generated from Duality AI's digital twin simulation platform, Falcon. The trained model must accurately segment the environment, a skill crucial for off-road autonomy in Unmanned Ground Vehicles (UGVs). The primary goal is to train the most accurate and precise model, evaluate its performance on a novel, but similar, desert environment, and benchmark and optimize it for real-world deployment.
The challenge emphasizes the use of synthetic data, which addresses the scarcity, cost, and time-consumption of traditional real-world data collection.
Judging Criteria (for Round 1 Evaluation):
Model Performance (80 Points): Measured primarily by the IoU Score.
Performance Report Clarity (20 Points): Based on structured findings, detailed reporting, and well-organized documentation of methodology, challenges, and solutions.
